referat 0.3.0

Nytt

• Codex (jobbkonto) kan nu väljas som provider för protokollsteget.
• Referat återanvänder en befintlig inloggad Codex CLI-session, så ingen OpenAI API-nyckel behöver lagras.
• Codex körs tillfälligt i en tom arbetskatalog med fil-, skal-, app-, plugin-, webb- och historikfunktioner avstängda.
• Inställningarna innehåller anslutningstest, svensk vägledning och en tydlig varning för möten klassade Highly Confidential.
• Befintliga lokala, OpenAI-, Azure-, Anthropic- och custom-providers fortsätter fungera som tidigare.

Codex CLI måste redan vara installerad och inloggad på datorn.

What's Changed

• Polish English README by @sockulags in #14
• Bump the github-actions group with 6 updates by @dependabot[bot] in #24
• Bump the lint-toolchain group across 1 directory with 2 updates by @dependabot[bot] in #27
• Bump the build-toolchain group across 1 directory with 2 updates by @dependabot[bot] in #28
• fix(e2e): fetch the Electron binary explicitly, and move to Node 24 by @sockulags in #29
• Add Codex CLI as a summary provider by @sockulags in #33
• Bump @types/node from 22.20.1 to 24.13.3 by @dependabot[bot] in #32
• Bump the react group with 2 updates by @dependabot[bot] in #31
• Release Referat 0.3.0 by @sockulags in #34

New Contributors

• @sockulags made their first contribution in #14
• @dependabot[bot] made their first contribution in #24

Full Changelog: v0.2.0...v0.3.0

referat 0.2.0

Speaker release: the transcript can now show who said what.

• Speaker identification (optional, off by default): a local companion server (diarization-server/, pyannote.audio) labels every transcript segment — Talare 1, Talare 2, … Click a label to rename it; regenerate the minutes and the names carry into the protocol. Setup guide: Speaker Diarization. IT can host one server for the whole office — users only enter an address.
• Voice recognition across meetings (opt-in): name a speaker once and later meetings suggest the name when the same voice is heard — always shown as a suggestion ("Anna?") that you confirm or dismiss, never applied silently. Voiceprints are stored locally only and can be deleted per voice or all at once. Off by default; see the privacy notes.
• Never blocks the minutes: if the speaker server is down, the meeting gets a notice and the protocol is still produced.
• One-shot server install script for Windows (install.ps1 + start-server.bat).
• Existing installs update automatically via the built-in updater.

Signed with the same self-signed certificate as v0.1.x (see v0.1.0 notes for the SmartScreen story).

referat 0.1.1

Maintenance-ready foundation release.

• Silent auto-updates: the app now checks GitHub Releases and updates itself quietly (installs on quit, or restart via the in-app toast). This is the first updatable version — from here on, patches arrive automatically.
• Committed test suite: 54 unit tests + full end-to-end walkthrough (onboarding → recording → pipeline → minutes), running in CI on every PR.
• Dependabot for npm and GitHub Actions.
• Signed with a self-signed certificate (see v0.1.0 notes for the SmartScreen story and the attached .cer for IT deployment).

referat 0.1.0

First public release.

• Record meetings (microphone + system audio) on Windows
• Automatic transcription and meeting minutes via configurable AI providers (local server, company server, OpenAI, Azure OpenAI, Anthropic)
• Everything stored locally; API keys encrypted with Windows DPAPI
• Export minutes as Word, Markdown or clipboard

Signing: this build is signed with a self-signed certificate, so Windows SmartScreen will still warn on first run (More info → Run anyway). Organizations piloting referat can remove the warning by deploying the attached referat-codesign.cer to Trusted Publishers via GPO/Intune. CA-issued signing is on the roadmap.